US AI stock sell-off shakes markets from Wall Street to Asia

As always, zoom out.

"South Korea’s benchmark closed 10% down on Tuesday after the country’s largest chipmakers, SK Hynix and Samsung Electronics, both closed over 12% lower. Japan’s Nikkei 225 was down 3.5% at the close of trading."

This is huge, isn't it? The conservative investors probably put their money into hardware companies, not directly into AI ones. So any scare of bubble-bursting affects memory manufacturing stock first and foremost? Meanwhile the AI stock itself continues to be held by the believers and Musk fanboys alike?

Why is the fear wondering if stocks going up immensely and then a little down important for HN? Since when are we talking stock prices here ?

All these media people who write about markets need to adjust their perspectives on what a sell-off means. A 2% reduction for a market cap of 80 trillion dollars still leaves about 78 trillion dollars in the market. The way the market behaves when there's 78 trillion dollars would be very different from how it would if there were 20 trillion dollars. With volumes of these kind, the instrument becomes a currency in itself causing a sort of runaway effect that will keep it going. When everyone's money is in the market, would the market ever fall?

> Elaborate.

In the dotcom era the demand was far lower than the projections, and orders of magnitude less than what they were building out to. The "products" like catsdotcom and dogsdotcom had massive investments with 0 real-world demand. Even hardware stuff like fiber rollouts were severely overestimated, and led to very cheap dark fiber for a while, because the demand simply wasn't there. We know now, with hindsight, that the demand eventually overtook those projections and deployments 10-15 years later.

In comparison, today we see huge demand, and all of the providers seem to be swamped. The hardware pricing has gone through the roof. Environment stuff + public "opinion" + outside influence campaigns have led to a very hostile environment for building out capacity. Real constrains for power generation. The providers are so eager to get more capacity online that even the most "religiously self proclaimed good guys" had to sign with spacemanbad to get more compute yesterday. It's that bad.

And despite all of this, all the providers are reporting increasing revenue. Even "wrappers" like cursor reported 4+B/y revenue. The top labs are remporting several B /mo in revenue, and growing. Even goog's cloud stuff is starting to be a sizeable chunk of their revenue, and that's competing with the ads cash cow. Hell, even meta had a record quarter, driven by increased traffic in their products, likely because they deployed chat stuff. There is real demand in the market today, and with model capabilities going up, that's bound to continue for the foreseeable future.

The dotcom bubble and this thing are nothing alike. Any of the large "products" can (and some probably will) fail, but the tech behind it is like the Internet. It will only grow, and be ubiquitous in a decade.